Yisca-git

Meal Mind Agent Yisca — Development skill for Claude Code

Development community

Meal-Mind: a meal & shopping planner agent built twice on the same tools, once on the raw Anthropic Messages API (hand-written harness) and once on the Claude Agent SDK, then measured and compared (to.

How to install Meal Mind Agent Yisca

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Yisca-git/Meal-Mind-Agent-Yisca and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Meal Mind Agent Yisca does

Alternatives in Development

Commerce Agents — Reference blueprint for building shopping and merchant agents with Claude 2.6k ★
Investigate Performance — Investigate a Nimbalyst performance problem (freeze, lag, idle CPU, slow op, memory growth) with measured evid 1.6k ★
Hive — Browser-native hive-mind for CLI coding agents — Claude Code, Codex, Gemini, and OpenCode collaborate as real 473 ★

README

Meal-Mind

A meal and shopping planner agent, built twice on the same tools and data and then measured:

API/ — the raw Anthropic Messages API. The whole harness (loop, tools, history, memory, context, compaction, caching, token metering) is hand-written in API/harness/, one file per responsibility.
SDK/ — the Claude Agent SDK. Claude Code runs the harness; this version only configures it. Every line that replaces a harness file is tagged # replaces: .
shared/ — the domain code both versions use: 8 tools (7 local, 1 external via TheMealDB), recipes, allergen/diet safety checks, the scenario runner and the pass/fail checks.

The agent remembers allergies, diet and servings between sessions, plans dinners, builds a shopping list minus the pantry, and refuses requests that contradict saved preferences (for example a pistachio cake with a nut allergy).

Results

Same scenario (7 turns, 2 sessions, 8 state-based checks), same model (`claude-opus-5-5`, effort `medium`), one run each:

	API	SDK
checks passed	8/8	8/8
model calls	22	28
cache hit	89.0%	93.5%
time	63.5 s	150 s
harness lines	206	27

The full comparison, with caveats, is in [REPORT.md](REPORT.md).

Setup

Python 3.13.

py -3 -m venv .venv
.venv\Scripts\python -m pip install -r requirements.txt

Run

# offline tests (no key needed)
.venv\Scripts\python -m pytest tests -q

# API version: needs ANTHROPIC_API_KEY set in the terminal only (never in code or git)
.venv\Scripts\python API\main.py --scenario scenario.json --out results\api.json

# SDK version: runs on your Claude Code login; ANTHROPIC_API_KEY is ignored for this process
.venv\Scripts\python SDK\main.py --scenario scenario.json --out results\sdk.json

Each run starts from a clean state directory (`.runs/`, git-ignored). Library versions are pinned in `requirements.txt`.